Hi Wookhee.
The first, though I find the notation odd. Factor graphs don't have
observed variables.
From the point of view of factor graphs, there are just factors, not
inputs, and so they are all the same model.
The way the features are interpreted are as drawn in the first one.
But you can easily implement the other two
by just changing the features.
Best,
Andy
On 06/18/2016 02:08 AM, Wookhee Min
wrote:
Dear Andreas,
I've found your PyStruct very useful to train CRF models.
Thanks very much for your work!
I have a quick question about the linear-chain CRF
implementation in PyStruct. Specifically, I am wondering which
of the three structures (in the figure below) the current
linear-chain
CRF implementation in PyStruct is grounded in. This
figure is from
slides
by Daniel Khashabi at UIUC.
Would you answer my question, when convenient?
Thank you.
Kind Regards,
Wookhee Min